Research Article
BeautyNet: Joint Multiscale CNN and Transfer Learning Method for Unconstrained Facial Beauty Prediction
Table 6
Prediction accuracy and Pearson’s correlation coefficient for transferring different layers.
| Transferring proposed network | 1 | 2 | 3 | 4 | 5 | Classification accuracy (%) | Pearson’s correlation coefficient |
| Scratch | 77.00 | 29.00 | 72.00 | 58.00 | 56.00 | 64.84 | 80.20 | Transferring conv1 | 66.00 | 39.00 | 75.00 | 56.00 | 55.00 | 65.92 | 79.00 | Transferring conv2 | 62.00 | 34.00 | 75.00 | 57.00 | 50.00 | 65.82 | 79.26 | Transferring conv3 | 77.00 | 30.00 | 74.00 | 52.00 | 53.00 | 65.52 | 79.37 | Transferring conv4 | 70.00 | 27.00 | 72.00 | 59.00 | 56.00 | 65.53 | 80.44 | Transferring conv5 | 66.00 | 40.00 | 71.00 | 61.00 | 58.00 | 65.92 | 81.07 | Transferring conv6 | 74.00 | 25.00 | 74.00 | 54.00 | 58.00 | 64.90 | 80.91 | Transferring conv7 | 69.00 | 36.00 | 73.00 | 57.00 | 55.00 | 66.02 | 81.46 | Transferring conv8 | 72.00 | 37.00 | 72.00 | 55.00 | 51.00 | 65.23 | 80.35 | Transferring conv9 | 70.00 | 25.00 | 80.00 | 62.00 | 39.00 | 65.82 | 81.59 | Transferring multiscale layer | 68.00 | 32.00 | 73.00 | 59.00 | 62.00 | 67.48 | 82.96 | Transferring Fc1 layer | 73.00 | 28.00 | 74.00 | 53.00 | 54.00 | 65.53 | 83.54 |
|
|
1, 2, 3, 4, and 5 show the classification accuracy for each specific category.
|